White Balance Correction Using Face Detection Reliability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional white balance gain calculation methods in digital cameras often incorrectly identify a person's complexion as white under tungsten light, leading to inaccurate white balance correction, especially in close-up shots, due to the similarity in color evaluation values between sunlight and tungsten light, and fail to handle cases where the face detection is erroneous or incomplete.
Innovation Solution
An image sensing apparatus that includes a face detection unit to identify and assess the reliability of face areas, and a white balance correction unit that adjusts the proportion of face area signals used in calculating white balance correction values based on the reliability of face detection, combining weighted white balance correction values from face and non-face areas to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the face area is removed from white detection target areas to prevent erroneous complexion correction, then the accuracy of white balance correction for non-face areas is improved, but the object areas for white detection disappear in close-up shots, causing a decrease in white balance correction accuracy
Solution Approach 1:
The patent applies local quality by differentiating the treatment of face areas and non-face areas. Face areas are excluded from automatic white balance correction to prevent complexion distortion, while non-face areas continue to provide white balance information. This selective approach maintains accurate white balance correction for the overall image while preventing erroneous correction of facial complexions.
Solution Approach 2:
The patent segments the image into face areas and non-face areas, applying different white balance correction strategies to each segment. The face detection unit identifies facial regions, and the white balance correction unit separately processes face and non-face areas, using non-face areas for white balance calculation while excluding face areas from automatic correction.
2Ease of manufacture
If conventional white balance gain calculation method is used, then the calculation process is simple, but complexion is erroneously judged to be white under tungsten light, causing inaccurate white balance correction
Solution Approach 1:
The patent introduces a face detection unit as an intermediary between the image sensing unit and the white balance correction unit. This intermediary identifies face areas and provides this information to the white balance correction unit, which then adjusts its calculation to exclude face areas from automatic white balance correction, preventing erroneous complexion correction while maintaining overall white balance accuracy.
Solution Approach 2:
The patent implements feedback by having the face detection unit continuously identify face areas and feed this information back to the white balance correction unit. The white balance correction unit uses this feedback to dynamically adjust which areas are included in white balance calculation, ensuring that face areas are excluded from automatic correction while maintaining accurate white balance for the rest of the image.
3Reliability
If face detection reliability is low or face areas are erroneously detected, then the white detection target area decreases, but the accuracy of output color temperature information declines
Solution Approach 1:
The patent changes the parameter of white balance correction by introducing a reliability threshold. When face detection reliability is below this threshold or when face areas are erroneously detected, the system adjusts the white balance correction parameters to exclude these unreliable face areas from the calculation, preventing degradation of color temperature information accuracy.
Data Source
AI summary
An image sensing apparatus that senses objects and processes image signals obtained from such image sensing includes a white balance correction means that performs white balance correction of the image signals using white balance correction values calculated based on a face area and on other areas contained in the image signals, and a CPU that changes the proportion of image signals of the face area used in calculating the white balance correction values when calculating the white balance correction values based on the image signals.